We have a quite accurate view of the X stack, so there's no good reason to ask
the X server to do restacking that has no effect. (Restackings that have no
effect on either X windows or Wayland windows were generally optimized out in
the synchronization code, but in other cases like moving an X window that is
only beneath Wayland windows to the top of the stack we would make such
requests.)
Removing such requests:
- Is a small efficiency win in itself
- Allows us to immediately go ahead and apply Wayland changes to the verified stack
- Prevents queued Wayland changes piling up waiting for an X event that will never
be received, since the X server will not send confirmation of no-op restacks.
Since such operations may still have an effect on the relative stacking of X
and Wayland windows, we need to continue applying them to the local stack.
https://bugzilla.gnome.org/show_bug.cgi?id=736559
Now that all actual stack shuffle is handled inside stack-tracker.c, we can make
meta_stack_tracker_record_[raise_above/lower_below] internal to that file and
remove the unused meta_stack_tracker_record_lower().
https://bugzilla.gnome.org/show_bug.cgi?id=736559
stack.c:sync_stack_to_xserver had both code for assembling the desired stack, and
code for enforcing the desired stack on the actual stack of X and Wayland windows;
the latter part is properly the domain of stack-tracker.c; moving the code to
apply the stack there both simplifies it and keeps stack.c more manageable.
https://bugzilla.gnome.org/show_bug.cgi?id=736559
There was still code in stack.c to handle skipping override-redirect windows,
but since quite a while ago, meta_stack_add() is not called for OR windows
since they are outside our stacking control. Add an assertion and remove
unnecessary code.
https://bugzilla.gnome.org/show_bug.cgi?id=736559
stack.c kept it's own record of the last stacking it requested, so that
restacking could be done with minimal moves, but we already have a better
view of the stacking order with the stack tracker, so use that instead.
This allows eliminating the special case for the first restack.
https://bugzilla.gnome.org/show_bug.cgi?id=736559
Since MetaStackTracker is the code that knows about the current X stacking order
and the relationship between X windows and Wayland windows, it's cleaner to
encapsulate stack manipulation in MetaStackTracker rather than have the calling
code make the X calls and only call into MetaStackTracker to inform it about
the changes.
https://bugzilla.gnome.org/show_bug.cgi?id=736559
The step where we requery the stacking order from the server than combine
it in an arbitrary fashion with Wayland windows can be eliminated by observing
that we are the final authority for Wayland window stacking - so if we
apply each X event that we receive from the X server to our stack in a
way that leaves the X windows ordered in the same way as on the server,
and apply events that we have stored locally in a way that doesn't affect
the ordering of X windows, than we have a fully correct ordering of windows.
Ordering this in the order of first applying the X event and then applying the
local portion also means that as long as we had an up-to-date view of the X
stack the composite operation will be identical to what was requested.
https://bugzilla.gnome.org/show_bug.cgi?id=736559
Putting X windows and pointers to MetaWindows into a union had a number of
problems:
- It caused awkward initialization and conditionalization
- There was no way to refer to Wayland windows (represented by
MetaWindow *) in the past, which is necessary for the MetaStackTracker
algorithms
- We never even cleaned up old MetaStackWindow so there could be
records in MetaStackWindow pointing to freed MetaWindow.
Replace MetaStackWindow with a 64-bit "stack ID" which is:
- The XID for X Windows
- a "window stamp" for Wayland windows - window stamps are assigned
for all MetaWindow and are unique across the life of the process.
https://bugzilla.gnome.org/show_bug.cgi?id=736559
Add a private hook for the test framework to get XSyncAlarmEvent events -
this will be used to implement XSyncCounter based synchronization
so that the test framework can deterministically wait until Mutter
has seen actions performed by an X11 client.
https://bugzilla.gnome.org/show_bug.cgi?id=736505
Setting the scaling factor immediately after calling clutter_init()
avoids creating the stage at one size, then later resizing it to
a different size.
https://bugzilla.gnome.org/show_bug.cgi?id=736279
We can enter weird states where get_default_window is called during
window unmanagement, before the window has been fully removed from
the stack. Make sure these windows are *never* returned from
get_default_window, as focusing them can cause an assertion fail,
or worse.
Clutter events include the layout index codified into modifier_state,
unlike XI2 device events, which means that we need to mask it out so
that we can match successfully.
A lot of applications assume that the window is fully positioned when it
gets the MapNotify, especially simple applications. Make sure that the
window is only mapped through the calc_showing logic.
O-R windows appear in workspace->windows, which aren't relocatable,
so we can't simply check if the workspace is empty after relocating
all normal windows, since those windows remain.
Make sure that the only windows we have are those that are
on_all_workspaces.
This is bad behavior, and can also cause us to get in an infinite loop
if an OR window is mapped and unmapped in quick succession. This
sequence causes a MapNotify followed by an UnmapNotify, and when
processing the events, we'll call XMapWindow, XUnmapWindow, which will
put another set of MapNotify, UnmapNotify events in our queue, which we
then process by calling XMapWindow, XUnmapWindow, and so it goes
forever, or at least some scheduler uncorks us by making us call
XMapWindow when the window is already mapped, or XUnmapWindow when the
window is already unmapped.
We can stop this madness by simply making sure never to call neither
XMapWindow or XUnmapWindow on OR windows, which is the correct thing to
do anyway.
We commonly used the generic, undetailed signal 'changed' to track
changes to preferences. Since we crash on unknown preference types,
this can be dangerous if somebody adds a new setting that has a
type we're unfamiliar with, and something else changes it.
Instead of crashing, just fizzle out doing nothing.
In gnome-shell, we have a feature where the user can unmaximize windows
by dragging them from the panel above the window. With accurate
anchoring, this looks really weird as the cursor is now "detached" from
the window. Detect this case and put the cursor in the middle of the
window titlebar instead.
This seems to be more accurate with what we currently see in
GNOME. Without having the app expose this information to us, it might be
a better idea to use the default frame style for this information instead.
The cursor was anchored wrongly when trying to unmaximize windows from
dragging them from the top of the screen because of a few think-o's and
some code that forgot to be updated when we moved to the frame rect
coordinate system.
The cursor is still offset for windows that start dragging from the top
panel. This is technically correct, but looks wrong. We'll fix this in
the next commit.
The initialization sequence before was quite icky, initializing Clutter
in a few different places depending on what was going on.
Put that all back into main.c
This removes our Xwayland dependency in the native path. The direct
grabs are still there for the X11 backend and are a bit disgusting,
but that's OK. We can refactor it out later.
This introduces some pretty lousy hackery because it depends on
https://github.com/xkbcommon/libxkbcommon/pull/10 , and I really
don't want to wait on that to squash this dep.
Now that the internal mutter bindings and gnome-shell stopped using
META_KEY_BINDING_REVERSES, and after moving the 'adding shift reverses
the keybinding action' logic to gnome-control-center, we can remove
META_KEY_BINDING_REVERSES from mutter.
Plugin API is broken as this constant is removed from the exported
headers. ABI is broken as using this flag is now a noop.
https://bugzilla.gnome.org/show_bug.cgi?id=732385
Currently the bindings for {switch,cycle}.* actions are created with the
META_KEY_BINDING_REVERSES flag so that <shift>+binding triggers the
reverse action. However, gnome-control-center does not know about this
kind of implicit bindings, and, for example, cannot warn when the user
tries to setup a conflicting <shift>+xxx binding.
These backward <shift> bindings are being explicitly set in
gsettings-desktop-schemas, so the META_KEY_BINDING_REVERSES annotation
can be removed for them from mutter.
https://bugzilla.gnome.org/show_bug.cgi?id=732385
MetaKeyBinding can be marked as being reversed
(META_KEY_BINDING_IS_REVERSED), but MetaKeyHandlerFunc callbacks
cannot check whether this flag was set or not on the MetaKeyBinding
which triggered the callback.
https://bugzilla.gnome.org/show_bug.cgi?id=732295
The existing workspace management code is quite hairy, with plenty of
logic inline in all of window.c, workspace.c, and screen.c, making it
hard to understand or make changes to, since you might forget to change
several of the other places the code was around.
Rewrite the internal workspace management logic so that it's
centralized and all in window.c. Document the invariants we need to
maintain, and ensure that these invariants are properly kept, with
asserts in various places.
Extensive testing on gnome-shell did not bring up any issues, and this
is a considerable cleanup.
This will be used to replace some of the hooks that are used to call
into window.c, so that the workspace index property is properly kept up
to date.
We can't name the property "index" since it causes conflicts with the
meta_workspace_index method. This should really be called
meta_workspace_get_index, but oh well.
I accidentally broke this in commit a119ea9. The code was considerably
more complicated than it needs to be, so let's replace it with a
g_list_find and nothing more.
Scanning over the hash table of XIDs is a terrible idea. Not only were
we excluding Wayland windows, but we were also looking at alarms and
barriers, too. We were lucky that that only contained GObjects where
our checks would work.
MetaGrabOp is painful and tedious to work with, because it's a
sequential series of values, meaning we have to use a giant unreadable
switch statement to figure out some basic things about the value.
To solve this, modify the encoding for MetaGrabOp and for the specific
window grab operations so that they're a set of bitflags that we can
easily check.
We've long used a switch statement on the grab operation to determine
where events should go. The issue with MetaGrabOp is that it's a mixture
of a few different things, including event routing, state management,
and the behavior to choose during operations.
This leads to poorly defined event routing and hard-to-follow logic,
since it's sometimes unclear what should point where, and our utility
methods for determining grab operations apart can be poorly named.
To fix this, establish the concept of a "event route", which describes
where events should be routed to.
It turns out that Clutter doesn't actually filter NumLock / ScrollLock /
CapsLock from button events due to its terrible event translation code.
Check only the grab mods to check if it's unmodified.
Popups could not set the cursor image, because the cursor tracker would
ignore window cursors if we had a popup active. The correct condition to
check for is already in should_block_wayland. Rename this to the more
sensible name windows_are_interactable, and use it in the cursor tracker.
Instead of returning a value based on whether or not we handled it, we
have this logic: either we have taken a grab on the window, in which
case we have a grab op and have handled it ourselves, or we did not take
a grab and *need* to replay the event to the window.
Handle this in events.c by checking the grab operation in the same way
that we check the other grab ops.
This is an accidental regression from 7a109a1. If we mark the event as
handled, then we *need* to set grab_op, or do some other sort of
behavior, since we have a grab.
Now that we always use XKB, it's very unlikely that we'll get a
MappingNotifier without a subsequence XkbKeymapNotify event. Just
do all the work always.
This will also help us considerably for the future when we'll be
putting the keymap event in the backend.
This allows creating the stage much earlier than it otherwise would have
been. Our initialization sequence has always been a bit haphazard, with
first the MetaBackend created, then the MetaDisplay, and inside of that,
the MetaScreen and MetaCompositor.
Refactor this out so that the MetaBackend creates the Clutter
stage. Besides the clarity of early initialization, we now have much
easier access to the stage, allowing us to use it for things such as
key focus and beyond.
Mutter depends on the X11 windowing backend of Clutter, unless it's used
as a Wayland display server.
This allows Mutter to run without breaking in case Clutter changes the
order with which windowing backends are selected, like it was the case
for bug https://bugzilla.gnome.org/show_bug.cgi?id=734587
The order of selection of the Clutter backends has not been made public,
so it cannot be relied upon since the introduction of the multiple
backends support; since Mutter requires the X11 backend functionality,
it should select the X11 windowing system, in the same way it selects
the EGL backend when compiled and run as a Wayland display server.
At this point there shouldn't be any system capable of running mutter
that doesn't have it and we're introducing functionality like setting
the keymap that has an hard requirement on it.
https://bugzilla.gnome.org/show_bug.cgi?id=734301
This way the xserver never paints the frame background, even if
the client window is destroyed. This allows us to have clean
destroy window animation.
There is no problem with interactive resizing because applications
are using the XSync protocol, so we're not painting unless the
client has redrawn.
https://bugzilla.gnome.org/show_bug.cgi?id=734054
MetaGestureTracker has been separating the "did I handle an event?" and the
"should the event be filtered out?" questions, merge this and make
handle_event() reply to "should the event be only handled by me?".
If a sequence wasn't accepted yet by the gesture tracker, the event will
go through (eg. not handled exclusively by the gesture tracker) and it'll
still be processed by Clutter, triggering gesture actions, and maybe
changing the sequence into other state.
https://bugzilla.gnome.org/show_bug.cgi?id=733631
On X11 this works because only emulated pointer events are listened for. On
wayland, the single touch behavior must be enforced in touch events, ignoring
every other sequence.
https://bugzilla.gnome.org/show_bug.cgi?id=733631
This function tells the obvious on X11, and implements a similar mechanism
on wayland to determine the "pointer emulating" sequence, or one to stick
with when implementing single-touch behavior.
https://bugzilla.gnome.org/show_bug.cgi?id=733631
Due to the way the MetaGestureTracker processes every touch event, this
will tell as closely to Clutter as possible the current number of touches
happening on the stage.
Even though, this is subject to windowing behavior, on X11, rejected touches
will be soon followed by a XI_TouchEnd event, so the compositor will stop
seeing touch sequences that are still operating on clients. On wayland, touch
sequences are processed by the compositor during all their lifetime, so these
will stay on the MetaGestureTracker with META_SEQUENCE_PENDING_END state, yet
still tracked.
https://bugzilla.gnome.org/show_bug.cgi?id=733631
On wayland, touches are initially both handled by the compositor and sent
to clients, proceeding to cancellation on clients only after the compositor
claims the sequence for itself. Implement the cancellation detail through
MetaGestureTracker::state-changed.
https://bugzilla.gnome.org/show_bug.cgi?id=733631
This reverts commit 3b85e4b2b9.
This breaks touch support; reverting would break wayland
(is what this patch tried to fix; we should find a better solution
that works on both).
The current GNOME Shell Alt-F2 restart looks very messy and also
provides no indication to the user what is going on. We need to
restart the compositor to switch in and out of stereo mode, so
add a framework for doing this more cleanly:
Additions:
meta_restart(): restarts the compositor with a message
MetaDisplay::show-restart-message: signal the embedding
shell to show a message
MetaDisplay::restart: signal the embedding shell to restart
itself.
meta_is_restart(): indicates whether the current instance is a
restart so we can suppress login animations.
A helper program meta-restart-helper holds the composite overlay
window up during the restart to avoid visual artifacts.
https://bugzilla.gnome.org/show_bug.cgi?id=733026
When a Wayland window acks our arrangement and we don't really have
anything to modify, we'll pass a sole flag of META_IS_WAYLAND_RESIZE
to meta_window_move_resize_internal using a garbage rect. The existing
code to calculate the new rectangle couldn't really handle this case,
and so the garbage rectangle accidentally got stored. Revamp the flag
checks to be more clear about it.
This fixes the weird positioning issues that sometimes appear when
resizing weston-terminal among others.
We really can't do this unless the backend X server is the same as the
frontend X server, as we pass a frontend XID to the backend, which is
only the case when we're not a Wayland compositor.
This code was supposed to refresh our default icons when the theme
changed, but it actually was a no-op, since the default icons are cached
in a static variable in MetaUI.
I'm not sure the fact that the fallback icons don't update when the
theme changes is an important enough use case to keep working, but I'm
keeping the skeleton function there in case somebody wants to actually
fix it properly.
Now that we have two connections to the X server, the idea of a
ref-counted server grab that might be held across extended portions
of code is very dangerous since we might try to use the backend
connection while the frontend connection is grabbed.
Replace the only usage (which was local) with direct
XGrabServer/XUngrabServer usage and remove the meta_display_grab()
API.
https://bugzilla.gnome.org/show_bug.cgi?id=733068
There's no obvious reason for grabbing the X server when unmanaging
a screen - the only race conditions a server grab solves are those
related to querying and then acting on the results of the query.
Our shutdown sequence is correctly ordered according to the ICCCM -
we first unselect on the root window, and then we destroy the
window owning WM_S<n> so removing the grab should not cause any
problems when we are being replaced with another window manager.
https://bugzilla.gnome.org/show_bug.cgi?id=733068
The only case we have is the case where the two X11 connections are the
same. When on Wayland, the XSync is costly and expensive, and we should
minimize it.
Commit 8100cefd4c fixed a crash during workspace initialization by
tweaking the startup sequence; as a result, the plugin (like gnome-shell)
is now started before workspaces are fully initialized, which breaks
some reasonable assumptions (like always having an active workspace).
This is particularly problematic considering that the code making those
assumptions is not necessarily our own (extensions!), so return to
fully initialize workspaces before the compositor again.
At the same time, make sure to only call meta_workspace_activate()
once during initialization to avoid reintroducing the crash.
https://bugzilla.gnome.org/show_bug.cgi?id=732695
This makes sure that we see them for Wayland clients as well, and don't
time out and crash when we're accessing an invalid window / surface.
Spotted-by: Rui Matos <tiagomatos@gmail.com>
If a sequence moves past a certain distance without being used by a
gesture, reject it so clients may see and react to it ASAP. This makes
gestures to be began by initially quasi-static touchpoints, in addition to
quasi-simultaneous.
Touch events will be caught first by the compositor this way,
whenever the MetaGestureTracker notifies of the accepted/rejected
state of a sequence, XIAllowTouchEvents() will be called on it
accordingly, so it is handled exclusively by the compositor or
punted to clients.
This object tracks both touch sequences happening on the stage and
gestures attached to the stage actor. When a gesture emits
::gesture-begin, All triggering sequences and future ones will be
marked as "accepted" by the compositor, and events will be listened
for meanwhile there are active gestures.
If a sequence goes unclaimed for a short time, it will be
automatically "denied", and punted to the client or shell element
below.
Touch events are largely ignored on GdkEvent emulation, so only
make frames receive pointer events, only the pointer emulating
touch will be reported, and any other further touches will be
ignored, which is about the behavior we want. This makes window
dragging possible again on touch.
Since Wayland configures are more of a hint to the client than anything,
we don't want to save the unconstrained rect when we're just hinting to
the client that it should resize, since it could ignore us. This would
get us stuck in a loop, since meta_window_move_resize_now would use the
unconstrained_rect to resize, and we don't remove the resize from the
queue if we have an outstanding request like that.
This fixes a bunch of traffic / CPU usage when trying to resize
weston-terminal.
For XWayland, we need to make sure to send out mouse events on O-R
windows, otherwise they won't get motion or button events.
The comment mentions being eaten for the compositor, but we already
bypass the compositor for all events that have a window. The return
value just controls whether we pass them to Wayland.
The output_id is more of an opaque identifier for the monitor, based on
its underlying ID from the windowing system. Since we also use the term
"output_id" for the output's index, rename our use of the opaque cookie
"output_id" to "winsys_id".
The GDK and hence GNOME standard is that keys that begin with XF86 according to
libxkbcommon not prefixed with XF86, though gdk_keyval_from_name() strips XF86
if provided. If libxkbcommon doesn't recognize the accelerator name without
XF86, try again adding XF86 to the start.
This restores compatibility with gnome-settings-daemon, schemas, and existing
user configuration.
https://bugzilla.gnome.org/show_bug.cgi?id=727993
It just gets in the way of gnome-shell's log handler (which
includes gjs backtraces optionally), it requires people to understand
what 8 or 16 mean as log levels, and it loses the log domain.
Some plugins and extensions want to be able to know when the sticky
field of a window changes, so add a property for it and allow them
to connect to the notify::on-all-workspaces signal.
When workspaces-only-on-primary is set and a window is moved back to the
primary, we also move it to the active workspace to avoid the confusion
of a visible window suddenly disappearing when crossing the monitor border.
However when the window is not actually moved by the user, preserving the
workspace makes more sense - we already do this in some cases (e.g. when
moving between primary monitors), but miss others (unplugging the previous
monitor); just add an explicit user_op parameter as used elsewhere to cover
all exceptions.
https://bugzilla.gnome.org/show_bug.cgi?id=731760
Remember the last monitor a window was moved to by user action and
try to move it back on monitor changes; this should match user
expectations much better when a monitor is unplugged temporarily.
https://bugzilla.gnome.org/show_bug.cgi?id=731760
When workspaces-only-on-primary is set, a window can be on all
workspaces either because it is on a non-primary workspace, or
because it was explicitly made sticky. Only the latter is reflected
in _NET_WM_STATE, but both will result in a "magic" _NET_WM_DESKTOP,
which we (and probably other WMs) use to set the initial sticky state.
So to avoid confusing other WMs (or ourselves), make sure to only
have _NET_WM_STATE_STICKY reflected in _NET_WM_DESKTOP when unmanaging.
Window state like maximization and minimization should be preserved
over restarts - in a patch review, this would qualify as "needs-work",
so revert the cleanup until the issues are fixed.
This reverts commit dc6decefb5.
Rather than calculate it speculatively with the current properties
which may be too new or too out of date, make sure it always fits
with the proper definition. We update it when we update the toplevel
window for X11, and when a Wayland surface is committed with a newly
attached buffer.
With get_input_region existing, get_input_rect is a misnomer. Really,
it's about the geometry of the output surface, and it's only used that
way in the compositor code.
Way back when in GNOME 3.2, get_input_rect was added when we added
invisible borders. get_outer_rect was always synonymous with server-side
geometry of the toplevel. get_outer_rect was used for both user-side
policy (the "frame rect") and to get the geometry of the window.
Invisible borders were meant to extend the input region of the frame
window silently. Since most users of get_outer_rect cared about the
frame rect, we kept that the same and added a new method, get_input_rect
to get the full rect of the framed window with all invisible borders for
input kept on.
As time went on and CSD and Wayland became a reality, the relationship
between the server-side geometry and the "frame rect" became more
complicated, as can be evidenced by the recent commits. Since clients
don't tend to be framed anymore, they set their own input region.
get_buffer_rect is also sort of a poor name, since X11 doesn't really
have buffers, but we don't really have many other alternatives.
This doesn't change any of the code, nor the meaning. It will always
refer to the rectangle where the toplevel should be placed.
All of the users of get_input_rect don't actually want a synthesized
input rect based off of the current margins. What they really want is
the last-configured size of the toplevel window.
Since we don't properly track this anymore in the generic MetaWindow,
use XGetWindowAttributes to fetch a server-side rectangle. This is a
bad layer violation, but since the window geometry code will have to
be rewritten anyway for the Wayland set_window_geometry, let's just
push a hacky fix for now.
Struts are defined in terms of screen edges, so expand the rectangles
we get via set_builtin_struts() accordingly. However we do want to
allow chrome on edges between monitors, in which case the expansion
would render an entire monitor unusable - don't expand the rectangles
in that case, which means we will only use them for constraining
windows but ignore them for the client-visible _NET_WORKAREA property.
https://bugzilla.gnome.org/show_bug.cgi?id=730527
Like the _NET_WM_STRUT/_NET_WM_STRUT_PARTIAL client properties,
_NET_WORKAREA is defined in terms of screen geometry rather than
taking individual monitors into account. However we do want to
allow system chrome to be attached to a monitor edge rather than
a screen edges under some circumstances. As not all clients can
be assumed to deal gracefully with the resulting workarea, use
those "struts" only internally for constraining windows, but
ignore them when exporting _NET_WORKAREA.
https://bugzilla.gnome.org/show_bug.cgi?id=730527
Since commit 8b2b65246a, we assume that the compositor always
exists. Alas, the assumption is wrong - the compositor is currently
initialized after the screen, but meta_screen_new() itself may
call a compositor function if initialization involves a workspace
switch (which will happen when meta_workspace_activate() is called
more than once and for different workspaces - or in other words,
when _NET_CURRENT_DESKTOP is set and not 0).
So carefully split out the offending bits and only call them after
the compositor has been initialized.
https://bugzilla.gnome.org/show_bug.cgi?id=731332
If we have a tree of a window, a non-attached dialog, and then an
attached dialog, we want to move the second window, not the attached
dialog or the topmost. In other words, we want to move the first
non-attached window, or the first "freefloating window".
This happens in Firefox, whose Preferences dialog is freefloating,
but suboptions of those are modal dialogs.
Stupid apps fullscreen themselves by resizing the client window to
monitor size. A monitor-sized frame rect on the other hand is perfectly
normal on monitors without struts - stop force-fullscreening those
and catch the real baddies instead.
https://bugzilla.gnome.org/show_bug.cgi?id=730681
When opening the window menu without an associated control - e.g.
by right-clicking the titlebar or by keyboard - using coordinates
for the menu position is appropriate. However when the menu is
associated with a window button, the expected behavior in the
shell can be implemented much easier with the full button geometry:
the menu will point to the center of the button's bottom edge
rather than align to the left/right side of the titlebar as it
does now, and the clickable area where a release event does not
dismiss the menu will match the actual clickable area in mutter.
So add an additional show_window_menu_for_rect() function and
use it when opening the menu from a button.
https://bugzilla.gnome.org/show_bug.cgi?id=731058
This can happen since we select for events on the root window, and
clients themselves might not select for input, meaning the X server
will bubble up. Just do nothing and ignore the event in this case.
This should hopefully fix some of the
Window manager warning: Log level 8: meta_window_raise: assertion '!window->override_redirect' failed
Window manager warning: Log level 8: meta_window_focus: assertion '!window->override_redirect' failed
spam that people have been seeing.
Since we often call meta_window_move_resize_now immediately after
mapping a window, we need to make sure that the placed coordinates
are saved in the unconstrained_rect. Ideally, placement positions
wouldn't be part of the constraints system, but instead are just
done inside meta_window_move_resize_internal as part of a special
path.
We're still working out the kinks of one large-scale refactor, so
it's best not to do another one while the first is going on. This
would be a great future cleanup, though: untangling constraints
and placement, alongside the force_placement state machine and
friends.
For Wayland, we want to have everything possible in terms of the frame
rect, or "window geometry" as the Wayland protocol calls it, in order
to properly eliminate some flashing when changing states to fullscreen
or similar.
For this, we need to heavily refactor how the code is structured, and
make it so that meta_window_move_resize_internal is specified in terms
of the frame rect coordinate space, and transforming all entry points
to meta_window_move_resize_internal.
This is a big commit that's hard to tear apart. I tried to split it
as best I can, but there's still just a large amount of changes that
need to happen at once.
Expect some regressions from this. Sorry for any temporary regression
that this might cause.
We have two different coordinate spaces here. One is the rectangle
returned by meta_window_get_frame_rect, which is called the "frame
rect" or "the window geometry", which includes visible frame borders
but not invisible frame borders. The other is "frame->rect" which
corresponds to the frame's server geometry. That is, it includes
both visible and invisible frame borders.
These two were of course the same until we introduced invisible
frame borders, and an executive decision was made to make
meta_window_get_frame_rect return the rectangle bounding the
visible portions of the frame.
As time went on, the "frame rect" turned out to be more useful when
making decisions upon, since the user often doesn't think about the
invisible window geometry as part of the window.
We already calculate what amounts to the "frame rect" in the theme
code, so just change META_CORE_GET_FRAME_RECT to consume that
directly.
Since we're going to be calling meta_window_get_frame_rect in here
soon, I'd rather it be one method call, rather than two. We can't
put it at the toplevel, since that might cause infinite recursion
(e.g. meta_core_get calls meta_window_get_frame_rect calls
meta_ui_get_frame_borders calls meta_core_get, ...)
Now that meta_window_move_resize and friends act in frame rect
coordinates, we need to convert the initial grab_anchor_window_pos
storage to be in frame rect coordinates as well.
This makes Alt+F7 / Alt+F8 work respectively under X11 nested mode.
For the native backend implementation, we'll need a special Clutter
function, so don't implement that for now.
The last commit added support for the "appmenu" button in decorations,
but didn't actually implement it. Add a new MetaWindowMenuType parameter
to the show_window_menu () functions and use it to ask the compositor
to display the app menu when the new button is activated.
https://bugzilla.gnome.org/show_bug.cgi?id=730752
We want to synchronize the button layouts of our server side
decorations and GTK+'s client side ones. However each currently
may contain buttons not supported by the other, which makes this
unnecessarily tricky.
So add support for a new "appmenu" button in the layout, to display
the fallback app menu in the decorations.
https://bugzilla.gnome.org/show_bug.cgi?id=730752
meta_window_get_position() returns the client rect position, which
we then pass to meta_window_move_frame. Just use the existing frame
rect coordinates.
The requested_rect is a strange name for it, because it's not actually
the rect that the user or client requested all the time: in the case of
a simple move or a simple resize, we calculate some of the fields
ourselves.
To the MetaWindow subclass implementations, it just means "the rect
before we constrained it", so just use the name unconstrained_rect.
This also makes it match the name of the MetaWindow field.
It looks weird to have Alt+Space pop up under the cursor instead
of the top-left corner of the window, and the Wayland request will
pass through the coordinates as well.
Add it to the compositor interface, and extend the
_GTK_SHOW_WINDOW_MENU ClientMessage to support it as well.
On X, basing the check whether the pointer is on the window on
Clutter events does not work, as the relevant events are handled
by GDK instead.
So add an X-specific window_has_pointer() implementation to also
fix mouse mode when running as X compositor.
https://bugzilla.gnome.org/show_bug.cgi?id=730541
Using clutter_actor_has_pointer() to test whether the pointer is
on the window makes for clean and nice-looking code, but does not
work in practice - ClutterActor:has-pointer is not recursive, so
we miss when the pointer is on the associated surface actor rather
than the actor itself.
Instead, check whether the window actor contains the core pointer's
pointer actor, which actually works.
https://bugzilla.gnome.org/show_bug.cgi?id=730541
Otherwise, the X server might read the backend's connection before
GTK+'s, meaning that it sees the XIGrabKeycode requests before the
CreateWindow.
This fixes keybindings on windows not working immediately.
Thanks to Rui Matos <tiagomatos@gmail.com> and
Julien Cristau <jcristau@debian.org> for helping track down the issue.
Realistically, the user rect contains the unconstrained window
rectangle coordinates that we want to be displaying, in case
something in the constraints change.
Rename it to the "unconstrained_rect", and change the code to always
save it, regardless of current state.
When metacity was originally being built, the purpose of the user
rect was a lot less clear. The code only saved it on user actions,
with various other calls to save_user_window_placement() and a force
mechanism sprinkled in to avoid windows being snapped back to odd
places when constraints changed.
This could lead to odd bugs. For instance, if the user uses some
extension which automatically tiles windows and didn't pass
user_action=TRUE, and then the struts changed, the window would be
placed back at the last place a user moved it to, rather than where
the window was tiled to.
The META_IS_USER_ACTION flag is still used in the constraints code
to determine whether we should allow shoving windows offscreen, so
we can't remove it completely, but we should think about splitting
out the constrainment policies it commands for a bit more
fine-grained control.
https://bugzilla.gnome.org/show_bug.cgi?id=726714
Now that we don't have to regrab to change the cursor, since it's
simply the cursor on the root window, all we have to do is update
the cursor on the screen.
We expect that meta_screen_set_cursor while grabbed will properly
set the cursor on the root window. Make sure this works by simply
always using the root cursor when we have an active grab.
It isn't necessary. As an X11 compositor, we'll only see the event
if we have the grab on the window, anyway.
This was causing issues moving windows as a Wayland compositor.
When we're a Wayland compositor, we get all the events, no exceptions,
so we don't need to grab.
This was masking focusing and raising issues under nested that showed
up under native.
If we apply a prediction immediately instead of queueing, we should
also free the operation immediately.
If we discard the prediction queue because we resync fully, we
need to free each operation too.
https://bugzilla.gnome.org/show_bug.cgi?id=729732
If we exit early as not handled, then the normal process_event
handler will fire, and trigger the overlay-key binding. As that's
a special binding that doesn't have a handler, trying to trigger
that handler will crash mutter.
Instead of returning early, just check for xdisplay every time
we try to drive the X grab state machine. We really need a better
solution for this on the Wayland side.
Since commit 6e8d1d79d, move operations are always performed for
the (toplevel) parent of all transient, which is just plain silly
if the dialog is not actually attached to its parent (either because
the dialog is not modal or the setting is disabled).
We need the old rect for two purposes: to find the x/y in a resize-only
action, and to pass into the constraints code for nefarious purposes.
The constraints code takes a frame rectangle, so we convert the original
client rect into a frame rect, but never convert it back since it's
unused for the rest of the function.
Instead of playing games with the variables, just have two,
separately-scoped variables. One is the client rect, the other is the
frame rect.
For gravity-based resizing, we need to make sure that the requested
rectangle has the proper x/y position given by the gravity resize,
rather than the bogus root_x_nw / root_y_nw parameter.
Make the test for this more explicit.
With our surface_mapped strategy, implement_showing might not
change whether the window has been shown or not, and thus we
might end up clearing pending_compositor_effect before the window
is mapped.
Only clear pending_compositor_effect when the effect has actually
been used.
Remove extend_by_frame and unextend_by_frame. Use a dumb hack in
window.c to translate into window geometry in back. We'll soon track
all rectangles in MetaWindow in terms of the window geometry.
Grab operations are now always taken on the backend connection, and
this breaks GTK+'s event handling.
Instead of taking a grab op, just do the handling ourselves. The
GTK+ connection will get an implicit grab, which means pointer /
keyboard events won't be sent to the rest of mutter, which is good.
Now that we grab devices on the X11 connection, we can run into
cross-connection issues. Since GTK+ frames are on the UI connection,
they'll get the passive grab when we click on them. Forcibly ungrab
on GTK+'s connection before attempting to take a grab on the backend
connection ourselves.
It's been long enough. We can mandate support for these, at least
at build-time. The code doesn't actually compile without either
of these, so just consider that unsupported.
Now that we have a global MetaScreen, we can simply have a global
MetaCursorTracker as well. Keep the get_for_screen() API around for
compatibility, though.
The Alt+F7 and Alt+F8 keybinds for moving and resizing windows allow you
to move and resize the window off the screen, so allow the same for the
menu items as well, since they're marked with the same accelerator.
https://bugzilla.gnome.org/show_bug.cgi?id=728617
This doesn't particularly matter, since we fall through into a default
case that does nothing right below, but this matches the other paths
and it prevents us from falling into a trap if we add other event types
below.
If we start a grab op from a keybind / menu, we'll handle the
ButtonPress and drop the grab then, never giving the window a chance
to handle what it needs to do before the grab is dropped.
This means that if you use Alt+F7 to move a window around, move it
to a side-tiling or maximization area, and then left-click, it will
just hang there in the sky.
The entire point of it was to check whether the window was on the
right screen. Since we don't handle multiple screens anymore, we
don't need to check anything anymore.
Looking at the code paths where is_mouse / is_keyboard are used,
all of them should never be run when dealing with a COMPOSITOR
grab op, since they're filtered out above or the method is just
never run during that time.
It's confusing that COMPOSITOR is in here, and requires us to
be funny with other places in code, so just take it out.
We track changes to windows fullscreen state and stacking order
to determine a monitor's in-fullscreen state, but missed the
obvious case of moving a fullscreen window between monitors.
https://bugzilla.gnome.org/show_bug.cgi?id=728395
Commit 585fdd781c not only removed the tabpopup, but set invalid
handlers (a.k.a. NULL) for those shortcuts; add back handling of
basic handling of those shortcuts by switching instantly without any
popups.
https://bugzilla.gnome.org/show_bug.cgi?id=728423
dx/dy should be against the regular window's rect, and need to
be ignored when we're resizing. Instead, we use gravity to anchor
the window's new rectangle when resizing.
Doing this synchronously means that zenity tries to initialize GTK+.
Under Wayland, that will try to connect back to mutter as a display
server. We're waiting on zenity to exit, and zenity is waiting for
a connection response. Deadlock.
Simply assume that zenity will support all the options we feed it,
since it should be the correct version. Perhaps we should replace
our use of zenity with a simple helper binary that we know will
have all the right options if this still isn't good enough.
Our focus stealing prevention is still mostly inherited from metacity;
in particular, a (non-transient) window that is not on the current
workspace will not be given focus. This behavior made sense in the
GNOME 2 days, where workspaces were separated much more strictly.
However this is no longer the case in GNOME 3 - activating a launcher
will switch workspaces if necessary, and so will the app switcher.
There is no good reason to not do the same for other user actions
like clicking a URL or activating a search result, so allow activation
of windows on non-active workspaces if a proper timestamp is supplied,
assuming that this is a strong enough indication that we are dealing
with a legitimate user action.
https://bugzilla.gnome.org/show_bug.cgi?id=728018
Effectively we have been accepting CurrentTime timestamps for years,
but still complained about "stupid pagers" when encountering them;
just accept that we will never limit treating 0 timestamps as current
time to pagers.
https://bugzilla.gnome.org/show_bug.cgi?id=728018
The code that restacks X11 windows at the end first tracks any
old windows we know about, and then handles any windows created.
It starts when it ended, and then walks forwards and then
back looking for the first X11 window it doesn't know about.
However, when there aren't any X11 windows, it flies off the end
of the array and starts looking through random memory.
When it finds the X11 window, it then goes through and then tries
to restack the remaining windows according to how we've sorted
them.
Unfortunately, META_WINDOW_CLIENT_TYPE_X11 is 0, which is quite
common in random memory we have lying around, so we enter that
path and then just crash.
Fix the buffer overrun by adding the proper bounds check to the
search.
You can easily reproduce this by opening a menu while bloatpad
is full-screen. Why it only crashes when full-screen and not
when a standard window, I have no idea.
The idea here is that while we take a WM-side grab, like a compositor
grab or a resizing grab, we need to remove the focus from the Wayland
client.
We make a special exception for CLICKING operations, because these
are really an internal state machine while you're pressing on a button
inside a frame, and in this case, we need to not kill the focus.
Except while reading _NET_WM_WINDOW_OPACITY, opacity is between 0 and 255. With
guint8, we'll get compiler warnings if arbitrary int values are passed.
https://bugzilla.gnome.org/show_bug.cgi?id=727874
A careful analysis of mutter's codebase shows that nothing actually
passes anything but 0 to this. gnome-shell has one instance, but it's
most likely a mistake.
Remove the grab_mask field and the one place in keybindings.c that uses it.
The parameter to begin_grab_op is left in for API compatibility reasons.
The make_toplevel / window_unmanaging interface has never made
a lot of sense to me. Replace it with set_window, which does
effectively the same thing.
It's still not perfect in the case of XWayland, but I don't think
XWayland will ever make me happy.
Right now this just has all of the files in one directory. We'll
be introducing more structure to this in the future, and build
a proper backend system.
This will allow us to have a MetaCursorReference 'subclass' that's
lazily loaded. We currently always load all the images.
The long-term plan is to have a subclass for each "backend" and only
have CoglTexture as a common denominator. For the nested X11 backend,
we use XDefineCursor on our stage window. For the Wayland backend, we
would use set_cursor on our stage surface. For the native backend, we
would use the GBM code that's there right now.
The CoglTexture is there to be a "shared fallback" between all devices,
and also for the get_sprite API.
The odd man out is the X11 compositor case. For that, we need to move
the responsibility of setting the final cursor image out of
MetaCursorTracker, and simply have it be about tracking the used sprite
image and pointer position.
We want to make this private, and have MetaCursorReference be
backend-defined, with the texture possibly loaded on demand.
We can't make the definition of MetaCursorReference truly private yet
because of the XFixes cursor. A victim of MetaCursorTracker trying to
do too many things at once...
I want the MetaCursorTracker to mostly be about retrieving cursor
information. Start moving the code that loads cursor images to a
new file, MetaCursor. Eventually, MetaCursorTracker's APIs will
all take MetaCursorReferences, and we can have a clean backend
split here.
For whatever reason, this hash table was in the generic
implementation section instead of the XSync implementation,
even though it's only used by the XSync implementation.
Use it as a first pass of things to move over.
The reason we don't simply use gdk_window_add_filter directly is
because of some twisted idea that any GDK symbol being used from
core/ is a layer violation. While we certainly want to keep any
serious GDK code out of ui/, event handling is quite important
to have in core/, so simply use a GDK event filter directly.
Really, visible_to_compositor means that the window is shown, e.g.
not minimized. We need to be using a boolean tracking whether we've
called meta_compositor_add_window / meta_compositor_remove_window.
This fixes a jump during window placement when a window appears.
visible_to_compositor should always be in sync with show_window /
hide_window calls, even when unmananging.
This fixes a crash where we call sync_window_state when the window
is unmanaging, since we use visible_to_compositor to determine whether
the compositor will crash.
This is actually wrong; we should be using the knowledge about
whether we have called add_window / remove_window. We'll introduce
this with a new boolean next time.
Compositors haven't been able to manage more than one screen for
quite a while. Merge MetaCompScreen into MetaCompositor, and update
the API to match.
We still keep MetaScreen in the public compositor API for compatibility
purposes.
We previously separated out MetaDisplay and MetaScreen. mutter
would only manage one screen, but we still kept a list of screens
for simplicity.
With Wayland support, we no longer care about the ability to
manage more than one screen at a time. Remove this by killing
the list of screens, in favor of having just one MetaScreen
in MetaDisplay.
We also kill off active_screen at the same time, since it's
not necessary anymore.
A future cleanup should merge MetaDisplay and MetaScreen. To avoid
breaking API, we should probably keep MetaScreen around as a dummy
type.
When I refactored this out into a vfunc, I forgot to change the
code that interprets the result flags to actually respect the
new FRAME_SHAPE_CHANGED result flag.
Since we weren't ever clearing the frame bounds, this meant that
the "shadow clip" wasn't ever updated as a result. Since right now
all Wayland surfaces are considered ARGB32, we always clip shadows
under frames, and thus shadows had this weird "punch-out" from the
first frame shape.
While the ICCCM mandates the use of this, it's not necessary under
a composited environment from my understanding, and it's a flat
out no-op under XWayland.
Looking at the other rootless servers like Xwin/Xquartz, it seems
that they contain code for colormap emulation, but they're actually
never used -- a bug prevents the code from ever being called. Given
that it's been this way since 2003, I'm going to hazard a guess that
not many apps using colormaps. Kill them off.
display.c is getting a bit crowded. Move most of the handling
out to another file, events.c.
The long-term goal is to have generic event handling here, with
backend-specific handling for the types of windows and such.
If we have a CLICKING grab op we still need to send events to xwayland
so that we get them back for gtk+ to process thus we can't steer
wayland input focus away from it.
https://bugzilla.gnome.org/show_bug.cgi?id=726123
This ensures that we send the proper leave and enter events to wayland
clients.
Particularly, this solves a bug in SSD xwayland windows where clicking
and dragging on the title bar to move the window only works on the odd
turn (unless the pointer moves away from the title bar between
tries). This happens because xwayland gets a button press but doesn't
see the release so when it gets the next button press it discards it
because its pointer button tracking logic says that the button is
already pressed. Sending the proper wayland pointer leave event fixes
it since wayland clients must forget about button state at that point.
https://bugzilla.gnome.org/show_bug.cgi?id=726123
